fs/xfs/xfs_dfrag.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_bmap_btree.h"
  27 #include "xfs_alloc_btree.h"
  28 #include "xfs_ialloc_btree.h"
  29 #include "xfs_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_inode_item.h"
  33 #include "xfs_bmap.h"
  34 #include "xfs_itable.h"
  35 #include "xfs_dfrag.h"
  36 #include "xfs_error.h"
  37 #include "xfs_vnodeops.h"
  38 #include "xfs_trace.h"
  39
  40
  41 static int xfs_swap_extents(
  42         xfs_inode_t     *ip,    /* target inode */
  43         xfs_inode_t     *tip,   /* tmp inode */
  44         xfs_swapext_t   *sxp);
  45
  46 /*
  47  * ioctl interface for swapext
  48  */
  49 int
  50 xfs_swapext(
  51         xfs_swapext_t   *sxp)
  52 {
  53         xfs_inode_t     *ip, *tip;
  54         struct fd       f, tmp;
  55         int             error = 0;
  56
  57         /* Pull information for the target fd */
  58         f = fdget((int)sxp->sx_fdtarget);
  59         if (!f.file) {
  60                 error = XFS_ERROR(EINVAL);
  61                 goto out;
  62         }
  63
  64         if (!(f.file->f_mode & FMODE_WRITE) ||
  65             !(f.file->f_mode & FMODE_READ) ||
  66             (f.file->f_flags & O_APPEND)) {
  67                 error = XFS_ERROR(EBADF);
  68                 goto out_put_file;
  69         }
  70
  71         tmp = fdget((int)sxp->sx_fdtmp);
  72         if (!tmp.file) {
  73                 error = XFS_ERROR(EINVAL);
  74                 goto out_put_file;
  75         }
  76
  77         if (!(tmp.file->f_mode & FMODE_WRITE) ||
  78             !(tmp.file->f_mode & FMODE_READ) ||
  79             (tmp.file->f_flags & O_APPEND)) {
  80                 error = XFS_ERROR(EBADF);
  81                 goto out_put_tmp_file;
  82         }
  83
  84         if (IS_SWAPFILE(file_inode(f.file)) ||
  85             IS_SWAPFILE(file_inode(tmp.file))) {
  86                 error = XFS_ERROR(EINVAL);
  87                 goto out_put_tmp_file;
  88         }
  89
  90         ip = XFS_I(file_inode(f.file));
  91         tip = XFS_I(file_inode(tmp.file));
  92
  93         if (ip->i_mount != tip->i_mount) {
  94                 error = XFS_ERROR(EINVAL);
  95                 goto out_put_tmp_file;
  96         }
  97
  98         if (ip->i_ino == tip->i_ino) {
  99                 error = XFS_ERROR(EINVAL);
 100                 goto out_put_tmp_file;
 101         }
 102
 103         if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 104                 error = XFS_ERROR(EIO);
 105                 goto out_put_tmp_file;
 106         }
 107
 108         error = xfs_swap_extents(ip, tip, sxp);
 109
 110  out_put_tmp_file:
 111         fdput(tmp);
 112  out_put_file:
 113         fdput(f);
 114  out:
 115         return error;
 116 }
 117
 118 /*
 119  * We need to check that the format of the data fork in the temporary inode is
 120  * valid for the target inode before doing the swap. This is not a problem with
 121  * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
 122  * data fork depending on the space the attribute fork is taking so we can get
 123  * invalid formats on the target inode.
 124  *
 125  * E.g. target has space for 7 extents in extent format, temp inode only has
 126  * space for 6.  If we defragment down to 7 extents, then the tmp format is a
 127  * btree, but when swapped it needs to be in extent format. Hence we can't just
 128  * blindly swap data forks on attr2 filesystems.
 129  *
 130  * Note that we check the swap in both directions so that we don't end up with
 131  * a corrupt temporary inode, either.
 132  *
 133  * Note that fixing the way xfs_fsr sets up the attribute fork in the source
 134  * inode will prevent this situation from occurring, so all we do here is
 135  * reject and log the attempt. basically we are putting the responsibility on
 136  * userspace to get this right.
 137  */
 138 static int
 139 xfs_swap_extents_check_format(
 140         xfs_inode_t     *ip,    /* target inode */
 141         xfs_inode_t     *tip)   /* tmp inode */
 142 {
 143
 144         /* Should never get a local format */
 145         if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
 146             tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 147                 return EINVAL;
 148
 149         /*
 150          * if the target inode has less extents that then temporary inode then
 151          * why did userspace call us?
 152          */
 153         if (ip->i_d.di_nextents < tip->i_d.di_nextents)
 154                 return EINVAL;
 155
 156         /*
 157          * if the target inode is in extent form and the temp inode is in btree
 158          * form then we will end up with the target inode in the wrong format
 159          * as we already know there are less extents in the temp inode.
 160          */
 161         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 162             tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
 163                 return EINVAL;
 164
 165         /* Check temp in extent form to max in target */
 166         if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 167             XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
 168                         XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
 169                 return EINVAL;
 170
 171         /* Check target in extent form to max in temp */
 172         if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 173             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
 174                         XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
 175                 return EINVAL;
 176
 177         /*
 178          * If we are in a btree format, check that the temp root block will fit
 179          * in the target and that it has enough extents to be in btree format
 180          * in the target.
 181          *
 182          * Note that we have to be careful to allow btree->extent conversions
 183          * (a common defrag case) which will occur when the temp inode is in
 184          * extent format...
 185          */
 186         if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
 187                 if (XFS_IFORK_BOFF(ip) &&
 188                     XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
 189                         return EINVAL;
 190                 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
 191                     XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
 192                         return EINVAL;
 193         }
 194
 195         /* Reciprocal target->temp btree format checks */
 196         if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
 197                 if (XFS_IFORK_BOFF(tip) &&
 198                     XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
 199                         return EINVAL;
 200                 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
 201                     XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
 202                         return EINVAL;
 203         }
 204
 205         return 0;
 206 }
 207
 208 static int
 209 xfs_swap_extents(
 210         xfs_inode_t     *ip,    /* target inode */
 211         xfs_inode_t     *tip,   /* tmp inode */
 212         xfs_swapext_t   *sxp)
 213 {
 214         xfs_mount_t     *mp = ip->i_mount;
 215         xfs_trans_t     *tp;
 216         xfs_bstat_t     *sbp = &sxp->sx_stat;
 217         xfs_ifork_t     *tempifp, *ifp, *tifp;
 218         int             src_log_flags, target_log_flags;
 219         int             error = 0;
 220         int             aforkblks = 0;
 221         int             taforkblks = 0;
 222         __uint64_t      tmp;
 223
 224         /*
 225          * We have no way of updating owner information in the BMBT blocks for
 226          * each inode on CRC enabled filesystems, so to avoid corrupting the
 227          * this metadata we simply don't allow extent swaps to occur.
 228          */
 229         if (xfs_sb_version_hascrc(&mp->m_sb))
 230                 return XFS_ERROR(EINVAL);
 231
 232         tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
 233         if (!tempifp) {
 234                 error = XFS_ERROR(ENOMEM);
 235                 goto out;
 236         }
 237
 238         /*
 239          * we have to do two separate lock calls here to keep lockdep
 240          * happy. If we try to get all the locks in one call, lock will
 241          * report false positives when we drop the ILOCK and regain them
 242          * below.
 243          */
 244         xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
 245         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 246
 247         /* Verify that both files have the same format */
 248         if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
 249                 error = XFS_ERROR(EINVAL);
 250                 goto out_unlock;
 251         }
 252
 253         /* Verify both files are either real-time or non-realtime */
 254         if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
 255                 error = XFS_ERROR(EINVAL);
 256                 goto out_unlock;
 257         }
 258
 259         error = -filemap_write_and_wait(VFS_I(tip)->i_mapping);
 260         if (error)
 261                 goto out_unlock;
 262         truncate_pagecache_range(VFS_I(tip), 0, -1);
 263
 264         /* Verify O_DIRECT for ftmp */
 265         if (VN_CACHED(VFS_I(tip)) != 0) {
 266                 error = XFS_ERROR(EINVAL);
 267                 goto out_unlock;
 268         }
 269
 270         /* Verify all data are being swapped */
 271         if (sxp->sx_offset != 0 ||
 272             sxp->sx_length != ip->i_d.di_size ||
 273             sxp->sx_length != tip->i_d.di_size) {
 274                 error = XFS_ERROR(EFAULT);
 275                 goto out_unlock;
 276         }
 277
 278         trace_xfs_swap_extent_before(ip, 0);
 279         trace_xfs_swap_extent_before(tip, 1);
 280
 281         /* check inode formats now that data is flushed */
 282         error = xfs_swap_extents_check_format(ip, tip);
 283         if (error) {
 284                 xfs_notice(mp,
 285                     "%s: inode 0x%llx format is incompatible for exchanging.",
 286                                 __func__, ip->i_ino);
 287                 goto out_unlock;
 288         }
 289
 290         /*
 291          * Compare the current change & modify times with that
 292          * passed in.  If they differ, we abort this swap.
 293          * This is the mechanism used to ensure the calling
 294          * process that the file was not changed out from
 295          * under it.
 296          */
 297         if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
 298             (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
 299             (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
 300             (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
 301                 error = XFS_ERROR(EBUSY);
 302                 goto out_unlock;
 303         }
 304
 305         /* We need to fail if the file is memory mapped.  Once we have tossed
 306          * all existing pages, the page fault will have no option
 307          * but to go to the filesystem for pages. By making the page fault call
 308          * vop_read (or write in the case of autogrow) they block on the iolock
 309          * until we have switched the extents.
 310          */
 311         if (VN_MAPPED(VFS_I(ip))) {
 312                 error = XFS_ERROR(EBUSY);
 313                 goto out_unlock;
 314         }
 315
 316         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 317         xfs_iunlock(tip, XFS_ILOCK_EXCL);
 318
 319         /*
 320          * There is a race condition here since we gave up the
 321          * ilock.  However, the data fork will not change since
 322          * we have the iolock (locked for truncation too) so we
 323          * are safe.  We don't really care if non-io related
 324          * fields change.
 325          */
 326         truncate_pagecache_range(VFS_I(ip), 0, -1);
 327
 328         tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
 329         if ((error = xfs_trans_reserve(tp, 0,
 330                                      XFS_ICHANGE_LOG_RES(mp), 0,
 331                                      0, 0))) {
 332                 xfs_iunlock(ip,  XFS_IOLOCK_EXCL);
 333                 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
 334                 xfs_trans_cancel(tp, 0);
 335                 goto out;
 336         }
 337         xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
 338
 339         /*
 340          * Count the number of extended attribute blocks
 341          */
 342         if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
 343              (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 344                 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
 345                 if (error)
 346                         goto out_trans_cancel;
 347         }
 348         if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
 349              (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
 350                 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
 351                         &taforkblks);
 352                 if (error)
 353                         goto out_trans_cancel;
 354         }
 355
 356         /*
 357          * Swap the data forks of the inodes
 358          */
 359         ifp = &ip->i_df;
 360         tifp = &tip->i_df;
 361         *tempifp = *ifp;        /* struct copy */
 362         *ifp = *tifp;           /* struct copy */
 363         *tifp = *tempifp;       /* struct copy */
 364
 365         /*
 366          * Fix the on-disk inode values
 367          */
 368         tmp = (__uint64_t)ip->i_d.di_nblocks;
 369         ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
 370         tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
 371
 372         tmp = (__uint64_t) ip->i_d.di_nextents;
 373         ip->i_d.di_nextents = tip->i_d.di_nextents;
 374         tip->i_d.di_nextents = tmp;
 375
 376         tmp = (__uint64_t) ip->i_d.di_format;
 377         ip->i_d.di_format = tip->i_d.di_format;
 378         tip->i_d.di_format = tmp;
 379
 380         /*
 381          * The extents in the source inode could still contain speculative
 382          * preallocation beyond EOF (e.g. the file is open but not modified
 383          * while defrag is in progress). In that case, we need to copy over the
 384          * number of delalloc blocks the data fork in the source inode is
 385          * tracking beyond EOF so that when the fork is truncated away when the
 386          * temporary inode is unlinked we don't underrun the i_delayed_blks
 387          * counter on that inode.
 388          */
 389         ASSERT(tip->i_delayed_blks == 0);
 390         tip->i_delayed_blks = ip->i_delayed_blks;
 391         ip->i_delayed_blks = 0;
 392
 393         src_log_flags = XFS_ILOG_CORE;
 394         switch (ip->i_d.di_format) {
 395         case XFS_DINODE_FMT_EXTENTS:
 396                 /* If the extents fit in the inode, fix the
 397                  * pointer.  Otherwise it's already NULL or
 398                  * pointing to the extent.
 399                  */
 400                 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 401                         ifp->if_u1.if_extents =
 402                                 ifp->if_u2.if_inline_ext;
 403                 }
 404                 src_log_flags |= XFS_ILOG_DEXT;
 405                 break;
 406         case XFS_DINODE_FMT_BTREE:
 407                 src_log_flags |= XFS_ILOG_DBROOT;
 408                 break;
 409         }
 410
 411         target_log_flags = XFS_ILOG_CORE;
 412         switch (tip->i_d.di_format) {
 413         case XFS_DINODE_FMT_EXTENTS:
 414                 /* If the extents fit in the inode, fix the
 415                  * pointer.  Otherwise it's already NULL or
 416                  * pointing to the extent.
 417                  */
 418                 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
 419                         tifp->if_u1.if_extents =
 420                                 tifp->if_u2.if_inline_ext;
 421                 }
 422                 target_log_flags |= XFS_ILOG_DEXT;
 423                 break;
 424         case XFS_DINODE_FMT_BTREE:
 425                 target_log_flags |= XFS_ILOG_DBROOT;
 426                 break;
 427         }
 428
 429
 430         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 431         xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 432
 433         xfs_trans_log_inode(tp, ip,  src_log_flags);
 434         xfs_trans_log_inode(tp, tip, target_log_flags);
 435
 436         /*
 437          * If this is a synchronous mount, make sure that the
 438          * transaction goes to disk before returning to the user.
 439          */
 440         if (mp->m_flags & XFS_MOUNT_WSYNC)
 441                 xfs_trans_set_sync(tp);
 442
 443         error = xfs_trans_commit(tp, 0);
 444
 445         trace_xfs_swap_extent_after(ip, 0);
 446         trace_xfs_swap_extent_after(tip, 1);
 447 out:
 448         kmem_free(tempifp);
 449         return error;
 450
 451 out_unlock:
 452         xfs_iunlock(ip,  XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 453         xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
 454         goto out;
 455
 456 out_trans_cancel:
 457         xfs_trans_cancel(tp, 0);
 458         goto out_unlock;
 459 }